HPE and Nvidia expanded the HPE AI Factory with Nvidia at HPE Discover Las Vegas on June 16, 2026, adding the Nvidia Vera CPU and the Nvidia Agent Toolkit to HPE Private Cloud AI and extending Nvidia Confidential Computing across the full portfolio. The HPE ProLiant Compute DL394 Gen12 with Vera will land in 2027, and the Vera Rubin NVL72 rack-scale system is now available from HPE for frontier-scale workloads above 1 trillion parameters. HPE is also bringing the HPE Compute XD700, built on Nvidia HGX Rubin NVL8, into the AI Factory with support for up to 128 Rubin GPUs per rack.
The pitch is that enterprises are moving agentic AI from proof of concept into production, and the underlying infrastructure has to change to keep up. Vera is positioned as the first CPU built specifically for agents, tuned for the tool calls, orchestration, and real-time data processing that long-running agent loops require. Pairing it with Rubin GPUs inside a turnkey Private Cloud AI stack is HPE's argument for why customers should not roll their own.
The New York Stock Exchange, working with Redpanda and HPE, is an early enterprise customer evaluating Vera CPU on the DL394 Gen12. That is a notable validation case: exchanges run latency-sensitive workloads where deterministic CPU behavior is non-negotiable, and putting an agent-tuned chip in front of market data pipelines is a real test of the architecture's claims.
“Vera is the first CPU built for agents — designed for the tool calls, orchestration and real-time data processing required across the agent loop — bringing deterministic, low-latency performance into HPE Private Cloud AI.”— Chris Marriott, Nvidia
Key facts
- 01HPE ProLiant Compute DL394 Gen12 with Nvidia Vera CPU ships in 2027 inside HPE Private Cloud AI.
- 02Vera Rubin NVL72 targets models above 1 trillion parameters, with HPE Compute XD700 racks supporting up to 128 Rubin GPUs.
- 03Nvidia Spectrum-X Ethernet delivers 1.6x higher networking performance for AI communication versus off-the-shelf Ethernet.
- 04NYSE, in collaboration with Redpanda and HPE, is an early customer evaluating the Vera CPU on DL394 Gen12.
- 05Nvidia Confidential Computing now spans HPE AI Factory at Scale, Sovereign AI Factory, and Private Cloud AI.
On the software side, the Nvidia Agent Toolkit arrives inside HPE Private Cloud AI bundled with Nvidia Nemotron open models, the Nvidia OpenShell secure runtime, and Nvidia NemoClaw blueprints. Together they form what HPE is calling an agentic AI operating system, with monitoring, governance enforcement, and tooling for building and running autonomous multi-agent systems. Private Cloud AI also adds secure local agent registration, requiring approval of models, skills, and tools against centralized policies before they execute.
HPE Zerto Software now detects rogue agent actions and uses continuous data protection to roll back to a clean state when an agent goes off the rails. HPE Alletra Storage MP X10000, which achieved foundation-level Nvidia-Certified Storage status, automatically applies metadata and governance policies to unstructured data destined for AI pipelines, which HPE says improves token throughput. The framing across both products is the same: agents will misbehave, and the storage and recovery layers need to assume that as a baseline.
Nvidia Confidential Computing is now available across HPE AI Factory at Scale, HPE Sovereign AI Factory, and HPE Private Cloud AI, delivered through HPE Services. The HPE ProLiant Compute DL380a is certified under the Nvidia-Certified Systems program for Confidential Computing, providing hardware-based protection for AI workloads and sensitive data while preserving Nvidia acceleration. Across the AI Factory solutions, Nvidia BlueField DPUs and Nvidia DOCA handle in-silicon zero-trust policy enforcement, runtime threat detection, and network encryption.
That matters for sovereign and on-premises deployments where customers are training models on proprietary data or running inference against regulated datasets. Confidential computing establishes a chain of trust via cryptographic attestation and encryption at every stage, which is increasingly a procurement requirement rather than a nice-to-have. Vera Rubin will ship with full-stack Confidential Computing across every chip.
On networking, every Vera Rubin NVL72 system will ship with Nvidia networking built in — Vera BlueField-4 DPUs, ConnectX-9 SuperNICs, and Spectrum-X Ethernet — with Spectrum-6 switching delivering 1.6x higher networking performance for AI communication versus off-the-shelf Ethernet. Spectrum-X Ethernet is now the standard across HPE AI Factory with Nvidia, including at-scale, sovereign, and turnkey configurations. HPE also confirmed Nvidia InfiniBand options announced at GTC in March, including Quantum-X800 InfiniBand paired with the HPE Cray Supercomputing GX5000.
The full-stack integration extends to HPE AI Factory at Scale, Sovereign AI Factory, and Private Cloud AI, all of which now ship with Nvidia RTX PRO 6000 Blackwell Server Edition GPUs, Spectrum-X Ethernet networking, BlueField-3 DPUs, and ConnectX-8 SuperNICs as configuration options. The Unleash AI partner program is also expanding at Discover with nearly a dozen new software partners, including Aizen, BridgeTEK, deepset, Deliverance, Faclon Labs, Gallop, Rocket, Supervity, Thales, Trustwise, and Vortiqx.
The open question is timing. Vera CPU availability slips to 2027, which gives competing CPU-plus-accelerator stacks — including AMD's MI series paired with Epyc and the various custom-silicon programs at hyperscalers — a real window to land enterprise agent workloads first. HPE and Nvidia are betting that the combination of an agent-tuned CPU, confidential computing, and a governance layer wins on procurement criteria that purely faster silicon does not address.
For Nvidia, the HPE expansion follows the company's MLPerf Training 6.0 sweep across all seven benchmarks earlier in the week, reinforcing the message that the Rubin generation is landing in enterprise channels with both performance and software stack credibility. The harder competitive question is whether HPE's turnkey AI Factory model — which sells the rack, the network, the governance, and the agent runtime as one SKU — pulls enterprise budget away from the build-it-yourself approach that still dominates among the most sophisticated buyers. If Vera and the Agent Toolkit deliver on the agent-reliability pitch in 2027, the integrated AI factory becomes the default purchase pattern for the second wave of enterprise AI adopters, and the channel economics of that shift accrue heavily to whoever owns the reference architecture.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
